Hierarchical Neural Coding for Controllable CAD Model Generation

This paper presents a novel generative model for Computer Aided Design (CAD)that 1) represents high-level design concepts of a CAD model as a three-levelhierarchical tree of neural codes, from global part arrangement down to localcurve geometry; and 2) controls the generation or completion of CAD models byspecifying the target design using a code tree. Concretely, a novel variant ofa vector quantized VAE with "masked skip connection" extracts design variationsas neural codebooks at three levels. Two-stage cascaded auto-regressivetransformers learn to generate code trees from incomplete CAD models and thencomplete CAD models following the intended design. Extensive experimentsdemonstrate superior performance on conventional tasks such as randomgeneration while enabling novel interaction capabilities on conditionalgeneration tasks. The code is available athttps://github.com/samxuxiang/hnc-cad.